Effects of spatial microstructure characteristics on mechanical properties of dual phase steel by inverse analysis and machine learning approach

This work aims to investigate complex relationship between microstructure characteristics and mechanical properties of dual phase (DP) steel through an inverse analysis based on Markov chain Monte Carlo (MCMC) method combined with meso-scale material modelling. In this framework, a machine learning approach as surrogate model was developed, in which support vector regression (SVR) and artificial neural network (ANN) were trained using results from representative volume element (RVE) simulations coupled with damage model, which were previously calibrated with experimental data of commercial DP steel grades. Moreover, specific microstructure descriptors including Moran’s index, martensite band index and martensite orientation were proposed for representing effects of spatial distributions of martensitic phase. As a result, inverse predictions of microstructure characteristics of DP steels for achieving defined yield strength, tensile strength, uniform elongation and toughness were presented. The inverse analysis could solve the non-uniqueness of structure–property relationships of steel, whereby significances of dispersed structures and aligned martensite bands were highlighted in details. The approach fairly dealt with multi-target optimization and high dimensional problem, which can be further applied as a guideline for designing DP microstructures with enhanced mechanical properties.